Common hard disk drives are storage devices comprising disks whose data-carrying surfaces are coated with a magnetic layer. Typically, the disks are positioned atop one another on a disk stack (platters) and rotate around an axis, or spindle. To store data, each disk surface is organized in a plurality of circular, concentric tracks. Groups of concentric tracks placed atop each other in the disk stack are called cylinders. Read/write heads, each containing a read element and a write element, are mounted on an actuator arm and are moved over the spinning disks to a selected track, where the data transfer occurs. The actuator arm is controlled by a hard disk controller, an internal logic responsible for read and write access. A hard disk drive can perform random read and write operations, meaning that small amounts of data are read and written at distributed locations on the various disk surfaces.
Each track on a disk surface is divided into sections, or segments, known as physical sectors. A physical sector, also referred to as a data block or sector data, typically stores a data unit of 512 bytes or 4 KB of user data.
A disk surface may be divided into zones. Zones are regions wherein each track comprises the same number of physical sectors. From the outside inward, the number of physical sectors per track may decrease from zone to zone. This approach is known as zone bit recording.
A computer, or host, accessing a hard disk drive may use logical block addresses (LBAs) in commands to read and write sector data without regard for the actual locations of the physical sectors on the disc surfaces. By means of a hard disk controller the logical block addresses (LBAs) can be mapped to physical block addresses (PBAs) representing the physical locations of sector data. Different mapping techniques for an indirect LBA-to-PBA read and write access are known in the prior art. In some embodiments LBA-to-PBA mapping does not change often. In other embodiments the LBA-to-PBA mapping may change with every write operation, the physical sectors being assigned dynamically.
The storage capacity of a hard disk drive can be increased, inter alia, by reducing the track pitch (i.e., track width) of the concentric tracks on the disk surfaces. This requires a decrease in the size of the read and write elements. However, without new storage technologies, a reduction in the size of the write elements is questionable, as the magnetic field that can be generated is otherwise too small to adequately magnetize the individual bits on the disk surface. A known solution is the shingled magnetic recording methodology, by which a write element writes data tracks in an overlapping fashion. Further information pertaining to shingled magnetic recording (SMR) can be found in U.S. Pat. No. 8,223,458 B2 and U.S. Pat. No. 8,432,633 B2, as well as in patent applications US2013/0170061 A1, US2007/0183071 A1 and US2012/0233432 A1.
With SMR, overlapping data tracks are grouped into bands, which are separated by inter-band gaps, also known as “guard bands,” “guard regions,” or “guard tracks.” Typically, to change the contents of a first track in an already populated band, it is necessary to read out and buffer all subsequent tracks of the band because after updating the data on that first track, rewriting the buffered data up to the next guard region is unavoidable as the wide write element will inevitably overwrite the data of each subsequent track. Due to the sequential and overlapping structure of SMR, even a small change to the contents stored in a band can result in a significant increase in the amount of data that must be read and written, thus leading to significant delays. Such a process is referred to as “read-modify-write” or “write amplification.”
Workloads such as databases often generate random write operations characterized by ongoing updates of small data blocks. These are the most expensive operations within an SMR storage system due to their significant write amplification, which negatively impacts performance. Moreover, increasing file and data fragmentation can slow an SMR hard disk drive much more than it can a conventional hard-disk drive. For these reasons, SMR hard disk drives are primarily intended for cold-storage applications, that is, for scenarios in which data are rarely altered. In the prior art SMR hard disk drives are deemed unsuitable as equal, universal substitutes for conventional hard disk drives.
Known solutions for reducing write-amplification have their disadvantages. One option is to buffer the data of incoming write commands and write the data in larger, contiguous blocks at a later stage. This only works as long as the average data throughput of the collected random write operations is sufficiently low. If the required data throughput is permanently too high for the low write performance of an SMR hard disk drive, even a large buffer will run over, leading to a drastic drop in performance. Furthermore, depending on the design, an additional and/or larger buffer, e.g., flash memory, can increase the production costs of an SMR hard disk drive.
Other known approaches for reducing write amplification include garbage collection, as is also used in solid state disks (SSDs). In contrast to conventional hard disk drives, the association between logical block addresses (LBAs) and physical block addresses (PBAs) is entirely mutable. A translation layer provides a link between LBAs and PBAs. The garbage collection may perform an internal “scrubbing” or other housekeeping tasks from time to time by moving data internally.
U.S. Pat. No. 7,443,625 B2, entitled “Magnetic disk drive,” describes a process that uses a “shift address table”. An internal “scrubbing” takes place at regular intervals during which the table is “cleaned up.” Patent application US2007/0174582 A1, entitled “Mutable association of a set of logical block addresses to a band of physical storage blocks,” describes how to reduce write amplification by means of mutable mapping between logical block addresses and physical sectors. During regular operation stored data can be moved to a different physical location, thereby changing the LBA-to-PBA association. A map or table is used to maintain the allocation or association status of each physical sector. The disclosure of this patent application is hereby incorporated by reference in its entirety.